Final Report Summary - CFD-DEM (Numerical simulation of sediment entrainment)
Sediment transport is important for predicting the impact of human intervention on river and coastal systems. In the European Union (EU), sediment-related river and coastal maintenance work costs many tens of millions of Euro per year and one of the barriers to reducing these costs is our present inability to accurately model sediment entrainment and transport processes. The main deficiency of current sediment transport models is that they do not take account of the turbulent structure of the flow and, more importantly, the interactions between the flow and the sediment. This project is aimed at investigating the influence of turbulence on the entrainment and movement of coarse particles on the bed of an open channel. We look at factors that influence the stability of the particles, such as drag and lift forces, degree of protrusion, blocking by particles upstream, effect of particles downstream on the points of rotation of a moving particle together with how the particles move and then become stationary again.
The work performed is as the follows:
(a) We have modified an in-house Computational fluid dynamics (CFD) compute programme called CgLes so that it can accommodate moving curved surfaces using the Immersed boundary (IB) method. A series of verification cases has been performed.
(b) We have also coupled CgLes and another in-house Discrete element modelling (DEM) compute programme called Y-code and the coupled programme can model the particles' movement, collision and also effects on surrounding flow. A series of verification cases has been performed.
(c) We have performed necessary modifications, optimisation and parallelisation of the coupled code to ensure that the combined code is working at its maximum efficiency.
(d) A Direct numerical simulation (DNS) of a fully developed turbulent open channel flow over and through a water-worked rough bed consisting of densely packed 2-3 layers of spheres have been performed. The Reynolds number of this simulation is much greater than that of our previous work.
(e) After a statistically steady flow had been obtained from (d), the particles were set free and their entrainment, drag and lift forces, and also surrounding turbulence structures were recorded. The influence of turbulence on the entrainment of the particles and the factors that influence the stability of the particles and the dynamic process of the particles' entrainment and movement were investigated.
(f) We have also performed a Large eddy simulation (LES) of sediment entrainment in turbulent open channel flow, same as the one in (e) but with a reduction of the resolution by a factor of 2 in all directions, to examine the suitability of LES in modelling particle entrainment and suspension. Good agreement with the DNS results has been achieved which provides concrete evidences of the fidelity of the LES in sediment transport simulation. This is extremely important work because considerable savings in CPU time can be made using LES instead of DNS.
(g) A series of LES of sediment entrainment in turbulent open channel flow with different Shield's numbers has been carried out to examine the influence of turbulent coherent structures on sediment transport in bed-load and suspended-load regimes.
(h) A DNS of sediment entrainment in turbulent open channel flow with doubled domain length and grid resolution, compared with the case of (e), has been carried out on the purpose of capturing both the largest and the smallest eddy structures in the channel. Totally, 1.6X10^9 grid points have been allocated to 1 200 Central processing unit (CPU)s, which makes the simulation the largest one of its kind.
The main results achieved so far are as the follows.
(a) Three major contributions which are an iterative IB method, a novel half-distributing forcing strategy and a new wall-layer model for LES have been made to the state-of-the-art of the IB method. A combined compute program which is capable of not only correctly modelling of the turbulence but also the movement of the particles has been developed.
(b) Numerical results of sediment entrainment in a turbulent channel flow are presented. Three DNSs with different values of the Shields Function S show three distinctly different sediment entrainment patterns. When S=0.045 (upper), just below the entrainment threshold of 0.055 particles are almost stationary. When S=0.065 (middle), bed load is observed. When S=0.5 (below), suspended load is observed.
(c) The sediment entrainment results clearly show a close relationship between the continuous particle movements and sweep events. It is found that pressure gradients play a much more crucial role in this dynamic process rather than the shear stress possibly due to the large grain-size adopted in this study. Moreover, particle entrainment tends to occur in a cluster manner and particles are more readily to be entrained at the downstream margin of the high pressure region caused by a sweep structure.
(d) The hydrodynamic forces acting on saltating particles and the particle collision process have been presented for the first time. The dominated continuous particle saltation shows evidence that a collision-rebounding process is possible. The strong correlation between the abrupt changes in particle stream-wise and vertical velocity indicates that the particle's upward momentum is transferred from the stream-wise momentum by particle-bed collisions. This shows a clearly different physics comparing with that for particle entrainment in which particle upward momentum is obtained from turbulent coherent structures.
(e) Four LESs with different values of the Shield's Function S show similar results on the normalised mean lift force on the particles, although the particle transport regimes of these simulations are quite different. The maximum of the mean lift force on the particles is about 0.6 times of the particles' submerged gravitational force.
(f) The mechanisms for the entrainment and the subsequent continuous saltation of large grain-size particles are different. Turbulence coherent structures, especially sweep structures, play a significant role in the sediment entrainment. However, for the subsequent continuous saltation, the collision parameters are crucial. As stated above, the gain in the upward momentum of a saltating particle is obtained from the loss of its stream-wise momentum at collision. So if there is not a significant collision between the particle and the rough bed (for example the particle hit the top leeside of a rough bed particle), the particle cannot obtain enough upward momentum to rebound from the rough bed and tends to slide on or roll over the bed particles. Because the particle cannot obtain enough stream-wise momentum due to the lower flow velocity in the vicinity of the bed, it will finally settle down. This particle will keep stationary until be entrained by a strong turbulence structure.
The potential benefits of the results are enormous as it is possible to investigate the exact particle movement mechanism - something that is very difficult, if not impossible, to adequately measure experimentally. From a geomorphological point of view, this study is a major contribution to the understanding of sediment dynamics by focusing on the motion of individual particles, turbulence around these discrete particles and the role of turbulent eddies. It will lead to improved understanding of bed mobility in rivers with non-cohesive sediment and finally a universal, accurate and practicable sediment transport model. It will also build up our knowledge on predicting the erosion of bed material around hydraulic structures, evaluating the influence of recently constructed dams on the ecosystem, estimating the effectiveness of the sediment-related river maintain work and so on.
The work performed is as the follows:
(a) We have modified an in-house Computational fluid dynamics (CFD) compute programme called CgLes so that it can accommodate moving curved surfaces using the Immersed boundary (IB) method. A series of verification cases has been performed.
(b) We have also coupled CgLes and another in-house Discrete element modelling (DEM) compute programme called Y-code and the coupled programme can model the particles' movement, collision and also effects on surrounding flow. A series of verification cases has been performed.
(c) We have performed necessary modifications, optimisation and parallelisation of the coupled code to ensure that the combined code is working at its maximum efficiency.
(d) A Direct numerical simulation (DNS) of a fully developed turbulent open channel flow over and through a water-worked rough bed consisting of densely packed 2-3 layers of spheres have been performed. The Reynolds number of this simulation is much greater than that of our previous work.
(e) After a statistically steady flow had been obtained from (d), the particles were set free and their entrainment, drag and lift forces, and also surrounding turbulence structures were recorded. The influence of turbulence on the entrainment of the particles and the factors that influence the stability of the particles and the dynamic process of the particles' entrainment and movement were investigated.
(f) We have also performed a Large eddy simulation (LES) of sediment entrainment in turbulent open channel flow, same as the one in (e) but with a reduction of the resolution by a factor of 2 in all directions, to examine the suitability of LES in modelling particle entrainment and suspension. Good agreement with the DNS results has been achieved which provides concrete evidences of the fidelity of the LES in sediment transport simulation. This is extremely important work because considerable savings in CPU time can be made using LES instead of DNS.
(g) A series of LES of sediment entrainment in turbulent open channel flow with different Shield's numbers has been carried out to examine the influence of turbulent coherent structures on sediment transport in bed-load and suspended-load regimes.
(h) A DNS of sediment entrainment in turbulent open channel flow with doubled domain length and grid resolution, compared with the case of (e), has been carried out on the purpose of capturing both the largest and the smallest eddy structures in the channel. Totally, 1.6X10^9 grid points have been allocated to 1 200 Central processing unit (CPU)s, which makes the simulation the largest one of its kind.
The main results achieved so far are as the follows.
(a) Three major contributions which are an iterative IB method, a novel half-distributing forcing strategy and a new wall-layer model for LES have been made to the state-of-the-art of the IB method. A combined compute program which is capable of not only correctly modelling of the turbulence but also the movement of the particles has been developed.
(b) Numerical results of sediment entrainment in a turbulent channel flow are presented. Three DNSs with different values of the Shields Function S show three distinctly different sediment entrainment patterns. When S=0.045 (upper), just below the entrainment threshold of 0.055 particles are almost stationary. When S=0.065 (middle), bed load is observed. When S=0.5 (below), suspended load is observed.
(c) The sediment entrainment results clearly show a close relationship between the continuous particle movements and sweep events. It is found that pressure gradients play a much more crucial role in this dynamic process rather than the shear stress possibly due to the large grain-size adopted in this study. Moreover, particle entrainment tends to occur in a cluster manner and particles are more readily to be entrained at the downstream margin of the high pressure region caused by a sweep structure.
(d) The hydrodynamic forces acting on saltating particles and the particle collision process have been presented for the first time. The dominated continuous particle saltation shows evidence that a collision-rebounding process is possible. The strong correlation between the abrupt changes in particle stream-wise and vertical velocity indicates that the particle's upward momentum is transferred from the stream-wise momentum by particle-bed collisions. This shows a clearly different physics comparing with that for particle entrainment in which particle upward momentum is obtained from turbulent coherent structures.
(e) Four LESs with different values of the Shield's Function S show similar results on the normalised mean lift force on the particles, although the particle transport regimes of these simulations are quite different. The maximum of the mean lift force on the particles is about 0.6 times of the particles' submerged gravitational force.
(f) The mechanisms for the entrainment and the subsequent continuous saltation of large grain-size particles are different. Turbulence coherent structures, especially sweep structures, play a significant role in the sediment entrainment. However, for the subsequent continuous saltation, the collision parameters are crucial. As stated above, the gain in the upward momentum of a saltating particle is obtained from the loss of its stream-wise momentum at collision. So if there is not a significant collision between the particle and the rough bed (for example the particle hit the top leeside of a rough bed particle), the particle cannot obtain enough upward momentum to rebound from the rough bed and tends to slide on or roll over the bed particles. Because the particle cannot obtain enough stream-wise momentum due to the lower flow velocity in the vicinity of the bed, it will finally settle down. This particle will keep stationary until be entrained by a strong turbulence structure.
The potential benefits of the results are enormous as it is possible to investigate the exact particle movement mechanism - something that is very difficult, if not impossible, to adequately measure experimentally. From a geomorphological point of view, this study is a major contribution to the understanding of sediment dynamics by focusing on the motion of individual particles, turbulence around these discrete particles and the role of turbulent eddies. It will lead to improved understanding of bed mobility in rivers with non-cohesive sediment and finally a universal, accurate and practicable sediment transport model. It will also build up our knowledge on predicting the erosion of bed material around hydraulic structures, evaluating the influence of recently constructed dams on the ecosystem, estimating the effectiveness of the sediment-related river maintain work and so on.
